SODAS Data Discussion 2 (Fall 2025)
Copenhagen Center for Social Data Science (SODAS) aspirers to be a resource for all students and researchers at the Faculty of Social Sciences. We therefore invite researchers across the faculty to present ongoing research projects, project applications or just a loose idea that relates to the subject of social data science.
Two researchers will present their work. The rules are simple: Short research presentations of ten minutes are followed by twenty minutes of debate. No papers will be circulated beforehand, and the presentations cannot be longer than five slides.
Discussion 1
Presenter: Frederik Hjorth
Title: The Political Origins of Critical Social Science
Abstract:
The interface between social science theories and politics is rarely studied systematically. Yet, there are many examples of political actors and movements shaping the social sciences. We study the emergence of "critical social science" as a political phenomenon, and investigate its political origins. Drawing on academic publication data, we propose a measurement strategy that captures the output of critical social science at the level of universities, subnational locations, and countries. Using our measure of critical social science, we chart its rise over time and space, and analyze its historical origins in radical student protest movements. Specifically, we look at whether the intensity and incidence of mass student protests 1968-1978 affects the university-level output of critical social science in the following decades. Our contribution highlights the importance of political history for understanding the trajectory of the social sciences, and how this can be studied rigorously in a large-n framework.
Discussion 2
Presenter: Hjalmar Bang Carlsen
Title: The Reliability of AI Interviewers
Abstract:
AI-interviewers, which employ large language models (LLMs) to conduct adaptive interviews, have been proposed as a hybrid method that combines the breadth of surveys with the depth of qualitative interviews. A key challenge for their use in comparative research is reliability: the extent to which AI-interviewers reproduce consistent question wording and order across interviews. This paper evaluates the reliability of different AI-interviewer designs, ranging from loosely guided agents that generate questions freely to constrained agents instructed to reproduce fixed wording and order. Using simulated interviews with synthetic respondents, we systematically test variation in question formulation, adherence to prescribed question order, and the ability to responsively update question wording and order correctly to the interview context. Our initial results show substantial unreliability, with unconstrained designs producing the greatest variability, including not only alternative phrasings but also entirely different questions that shift the construct being measured. Even in more constrained designs, deviations in order and wording remain frequent and substantively consequential. We conclude that while responsiveness is a strength of AI-interviews, preserving comparability requires stricter design controls to limit interviewer-induced variability without eliminating adaptive probing.